Elim and Baseelim - Staff please take notice

[qan]

someone messed with the rating system and now soz3 is #1. who wants to play that when someone so bad could be at the top with the ratings... Fix the ratings

Here's the issue. Anyone who plays much more frequently will, by nature of Confidence, be able to be fairly high on the ladder. Maybe confidence is given a bit too much sway at present, and needs to be taken back a bit. Comments welcome. The other extreme is what we used to have -- once you had enough k+d to ladder (about 15-40 games), you could potentially have the #1 score, and quit as the season's elim champion. That was much more demotivating, I think.


At present the bot starts you at 20% confidence. At this baseline level of confidence, your ladder (adjusted) rating will be equal to 1/5 of your unadjusted rating, because ladder rating is just rating multiplied by confidence, or in this case, rating * .20 Currently, each game played adds .2% confidence, so every 5 games you add 1%.


Let's look at some examples using various unadjusted ratings (ratings are in the 1st row). Base unadjusted rating is 300, FYI. You should have a 300 rating if you average a 1:1 KDR, and are mostly killing other players who have a 1:1.

						Rating
Games	Confidence	50	100	150	200	250	300	350	400	450	500
10	22%	11	22	33	44	55	66	77	88	99	110
50	30%	15	30	45	60	75	90	105	120	135	150
100	40%	20	40	60	80	100	120	140	160	180	200
150	50%	25	50	75	100	125	150	175	200	225	250
200	60%	30	60	90	120	150	180	210	240	270	300
250	70%	35	70	105	140	175	210	245	280	315	350
300	80%	40	80	120	160	200	240	280	320	360	400
350	90%	45	90	135	180	225	270	315	360	405	450
400	100%	50	100	150	200	250	300	350	400	450	500

As you can see here, if you have played 400 or more games, you will have a big advantage over others in terms of ladder. Even a mediocre player with 200 unadjusted rating but 400 games played will be laddered equal to someone with 400 rating but only 150 games played.

[qan]

Here are some alternate scenarios. Take a look at the data and think about where you want players to be in terms of laddered rating vs number of games played. The point of Confidence is to ensure a player has enough games under their belt that the rating is a true indicator of their skill, and not some fluke.

Base confidence: 40%
100% confidence at: 300 games
Confidence increase per game: 0.2%

						Rating
Games	Confidence	50	100	150	200	250	300	350	400	450	500
10	42%	21	42	63	84	105	126	147	168	189	210
50	50%	25	50	75	100	125	150	175	200	225	250
100	60%	30	60	90	120	150	180	210	240	270	300
150	70%	35	70	105	140	175	210	245	280	315	350
200	80%	40	80	120	160	200	240	280	320	360	400
250	90%	45	90	135	180	225	270	315	360	405	450
300	100%	50	100	150	200	250	300	350	400	450	500

Base confidence: 29%
100% confidence at: 250 games
Confidence increase per game: 0.295%

Rating Games Confidence 50 100 150 200 250 300 350 400 450 500 10 32% 16 32 48 64 80 96 112 127 143 159 50 43% 22 43 65 87 108 130 152 173 195 216 100 58% 29 58 86 115 144 173 202 230 259 288 150 72% 36 72 108 144 180 216 252 287 323 359 200 86% 43 86 129 172 215 258 302 345 388 431 250 100% 50 100 150 200 250 300 350 400 450 500

This last one is I think my favorite -- it gives lower-than-average players a chance to get up the ladder a bit by playing a lot of games, while at the same time prevents them from topping it. Meanwhile, even an excellent player with a 500 rating will need to play a fair number of games to start ranking quite highly. But in the meantime they'll still be ranked higher than most players. As long as they perform consistently they will continue to increase in rank until they near the top, providing a feeling of overcoming.


If you guys would like to see how different variables play out, post them here and I can crunch them pretty easily.

[Spacely]

I have a suggestion that could cross check your confidence thing and possibly improve the ladder

yahoo games used to have ladders and rating the ladders was a special room so in regular rooms every game you add/lose rating in the ladder rooms you would also add/lose rating but along side that you would have rung calculated separately from rating so if you coded a rung system and added it along side the rating system you could see how they matched up

some basics on the yahoo ladder system

the most you can lose in any game is 1 rung unless you on the bottom rung. you go halfway between you and the rung you beat unless lower then you. so if rung 100 beat rung 1 then 100 would become 50 and 1 would become 2. not sure exactly but if you only played people lower then you after 3 to 5 wins you would gain a rung unless in top 20, 10, 5, 4, 3, 2, or 1. they all had different requirements to gain a rung. if you didn't defend your rung by playing every 2 days you would lose 1 rung

I like rung systems much better then rating systems. losers are easy because everyone loses 1 rung. if you can figure out a formula to calculate rung gained for winner vs multiple opponents this would be a really good system even if you only used it to run alongside your confidence system and gauge it

maybe take top 3 or 4 rungs played that round and average them to calculate the rung or something I don't know.

thank you for your time

[Spacely]

Here are some alternate scenarios. Take a look at the data and think about where you want players to be in terms of laddered rating vs number of games played. The point of Confidence is to ensure a player has enough games under their belt that the rating is a true indicator of their skill, and not some fluke.

Base confidence: 40%
100% confidence at: 300 games
Confidence increase per game: 0.2%

						Rating
Games	Confidence	50	100	150	200	250	300	350	400	450	500
10	42%	21	42	63	84	105	126	147	168	189	210
50	50%	25	50	75	100	125	150	175	200	225	250
100	60%	30	60	90	120	150	180	210	240	270	300
150	70%	35	70	105	140	175	210	245	280	315	350
200	80%	40	80	120	160	200	240	280	320	360	400
250	90%	45	90	135	180	225	270	315	360	405	450
300	100%	50	100	150	200	250	300	350	400	450	500

Base confidence: 29%
100% confidence at: 250 games
Confidence increase per game: 0.295%

Rating Games Confidence 50 100 150 200 250 300 350 400 450 500 10 32% 16 32 48 64 80 96 112 127 143 159 50 43% 22 43 65 87 108 130 152 173 195 216 100 58% 29 58 86 115 144 173 202 230 259 288 150 72% 36 72 108 144 180 216 252 287 323 359 200 86% 43 86 129 172 215 258 302 345 388 431 250 100% 50 100 150 200 250 300 350 400 450 500

This last one is I think my favorite -- it gives lower-than-average players a chance to get up the ladder a bit by playing a lot of games, while at the same time prevents them from topping it. Meanwhile, even an excellent player with a 500 rating will need to play a fair number of games to start ranking quite highly. But in the meantime they'll still be ranked higher than most players. As long as they perform consistently they will continue to increase in rank until they near the top, providing a feeling of overcoming.


If you guys would like to see how different variables play out, post them here and I can crunch them pretty easily.

so after players hit 250 games confidence isn't a factor anymore??(if so wouldn't it be better if you just gained .02% confidence per game but could also lose some confidence for certain things like losing to fast if you lasted like only 25% of the total game time you lose some confidence and things like that if a 10% won that round you would lose confidence b ut might also make it to complicated)

does confidence effect points lost? or is it always 100% of points lost in a round?

if so and I am at 32% and lose I would have to win at least 3 games to gain back the 100% I just lost?

[qan]

That's right, at a certain number of games, confidence is no longer a factor -- it's reached 100%, full confidence, because we're saying we totally believe that the player is of the skill level that their rating (based on overall K/D ratio and the average rating of the people they kill). The more games you've played, the more data is available of your skill level, and so the more confident we can be about your rating.


Confidence doesn't affect points lost, really, not in any direct way. Having a greater confidence value is always good for you. It's a direct multplier to your rating.


Rather than losing confidence, we could definitely do something to affect rating in certain situations, such as a fast loss, a dominating win, etc. However, a lot of those things are tricky to accurately gauge, and some could be gamed in order to cheat the system. The rung idea sounds pretty interesting. As you mention it might be hard to implement with many players involved, though. The idea that you lose a rung when you don't play for 2 days is an interesting idea. We could reduce rating slightly after a certain time interval of play, to make it more difficult for players to hold on to a top spot.

[qan]

OK, found what we need. Not sure why I overlooked this before... I'm going to make confidence logarithmic. Confidence = log2(rounds) / 10

Basically what this means is that it will increase very rapidly at the start, say for the first 50 games played, which is enough to give a pretty strong amount of confidence. But it will increase more and more slowly with each game played. This accurately reflects how it's supposed to work. For example, if you have a player with 300 rating with 700 games played vs and one with a 300 rating and 800 games played, everyone is about as confident that 300 is the real rating for both. But the difference in perceived confidence in a 300 rating for a player at 50 games vs one who's played 150 is much greater.

Doing it this way allows confidence to continue to rise the longer you play, which, if you play pretty consistently, will continue to increase your ladder rating slightly over time. But it also lets players who just began playing elim get a fairer ladder rating early on.

Games	Confidence	50	100	150	200	250	300	350	400	450	500
10	33%	17	33	50	66	83	100	116	133	149	166
20	43%	22	43	65	86	108	130	151	173	194	216
30	49%	25	49	74	98	123	147	172	196	221	245
40	53%	27	53	80	106	133	160	186	213	239	266
50	56%	28	56	85	113	141	169	198	226	254	282
100	66%	33	66	100	133	166	199	233	266	299	332
125	70%	35	70	104	139	174	209	244	279	313	348
150	72%	36	72	108	145	181	217	253	289	325	361
200	76%	38	76	115	153	191	229	268	306	344	382
250	80%	40	80	119	159	199	239	279	319	358	398
300	82%	41	82	123	165	206	247	288	329	370	411
350	85%	42	85	127	169	211	254	296	338	380	423
400	86%	43	86	130	173	216	259	303	346	389	432
450	88%	44	88	132	176	220	264	308	353	397	441
500	90%	45	90	134	179	224	269	314	359	403	448
550	91%	46	91	137	182	228	273	319	364	410	455
600	92%	46	92	138	185	231	277	323	369	415	461
650	93%	47	93	140	187	234	280	327	374	420	467
700	95%	47	95	142	189	236	284	331	378	425	473
750	96%	48	96	143	191	239	287	334	382	430	478
800	96%	48	96	145	193	241	289	338	386	434	482
850	97%	49	97	146	195	243	292	341	389	438	487
900	98%	49	98	147	196	245	294	343	393	442	491
950	99%	49	99	148	198	247	297	346	396	445	495
1000	100%	50	100	149	199	249	299	349	399	448	498

[qan]

OK, so I updated the algorithm tonight to use the above. You'll notice your ladder position may have changed a bit. Let me know what you think. Didn't have a chance to test it on the bot but I stared at it quite a few times and it didn't squirm so it should be legit.

[Spacely]

qan not to bicker but those numbers you have charted out did you make some line of code to chart them out because I noticed something messed up kinda if want to be picky like

33% of 250 is 82.5 but rounds up to 83 is ok
33% of 500 is 165 but for some reason it round up to 166
and then
66% of 500 is 330 but for some reason it rounds up 2 points to 332

simple fix maybe add decimals to the ranking points?

so not sure what is going maybe on with your coding or if you took a code snippet from somewhere but the 1/3 shouldn't round up I could see the 2/3 possibly rounding up 1 cause its more then 50% but I don't know to much about coding or how much this will effect things on the other end when you checking puts but just figured I would point this numbers thing out to you my friend

thank you for your time

[qan]

Ah, yeah, that's just a function of the spreadsheet. I can't remember what I had it doing, I didn't save it ... I remember seeing it too, though (clearly 498's not 100% of 500 for example).

Basically the actual percentages are what's rounded, iirc, not the resulting numbers. So 100% is actually 99.6%. log2(1000) is .996 ... but that's pretty damn close to 100, which was what I was going for. Had to play around a bit to find a desirable curve that also hit close to a nice round (and large) number.

Bottom line is that it should be safe. 1000 rounds played is max confidence, but in reality it's slightly less than 1004 games, by the formula ... though, in the code, once you hit 1000, it just levels it off at 1000 anyway.

Roundabout explanation but I hope that assuages your fears. I'm a little drunk right now so I apologize if the explanation is lacking